You may have seen the news last week that Gallup will no longer track presidential approval after 88 years.
Here’s a timeline:
- 1938 – 2025: Gallup tracked presidential approval (The Guardian)
Most recently, it was asked within the GPSS (Gallup Poll Social Series) - 1936 – 2012: Gallup polled presidential vote choice (Politico, this blog)
(The ijsbeer exploring a country with effectively 7.03 parties.)
Gallup reported Trump’s approval rating at 36% in December. This is lower than Nate Silver’s numbers:
It is difficult to assess Gallup’s accuracy because:
- Vote choice can be compared to election results to assess a pollster’s accuracy. But Gallup doesn’t poll vote choice anymore.
- I can’t find the 2025 GPSS response rate (in 1997 it was 28% and in 2017 it was 7%)
Let’s look at Gallup’s GPSS sampling methodology:
dual-frame design, which includes both landline and cellphone numbers. Gallup samples landline and cellphone numbers using random-digit-dial methods…
Their estimation uses weights:
Gallup weights samples to correct for unequal selection probability, nonresponse, and double coverage of landline and cellphone users in the two sampling frames.
They calibrate these weights to population targets:
Gallup also weights its final samples to match the U.S. population according to gender, age, race, Hispanic ethnicity, education, region, population density, and phone status…
Do you think this enough to correct for nonresponse bias ?
In 2024, Gallup’s director of U.S. survey research wrote:
Gallup is committed to maintaining its trends for as long as possible.
:(
To maintain trends, they have been keeping mode the same:
But the question of “how” is before us….Telephone surveys are becoming harder and more expensive… most surveys across the industry to move to self-administered modes, such as web and mail…“mode effect”…These effects would make it difficult to attribute…true shifts in opinion versus artifacts of a switch in methodology.
But there is a trade-off between that concern and nonresponse bias, which differs across modes and is worsening:
…Changes in respondent device use and call-blocking technology have led to declining contact rates…
Polls that attempt to capture voter sentiment are calibrated once there is an election, so at least every now and then you get some information about how well your adjustments are doing. But more general “presidential approval”-type polls… how can we ever know they actually capture the sentiment of “the American people” or even likely voters or whatever? I note that even if I disapprove of the job the president is doing, I might still vote for him over a challenger I think is even worse.
All of that said, I think it’s a pity that Gallup is stopping this poll. The fact that it’s getting harder to do a good job because the response rate keeps decreasing…I get that, but my gut feeling (not backed up by deep thinking, I admit) is that we are going to end up with less ability to figure out the mood of the country when it comes to presidential behavior and that that’s a bad thing. It’s funny, thanks to social media there is more information than ever about what people think — tens of millions of people post things publicly every day — but when it comes to getting something like the central tendency of people’s preferences I feel like we may be worse off than we were thirty years ago… although this perception may be completely wrong.
Thanks, Phil !
Yes, vote choice can be compared to election results to assess a pollster’s accuracy. Other questions never get to see such a ground truth.
And yes I agree that even with it being hard to assess the accuracy of the presidential approval polling, it seems bad to lose the time series. I agree that social media data isn’t easily translated to a polling result.